Merge branch 'docs' of git://git.lwn.net/linux-2.6
[linux-2.6] / drivers / net / enic / enic_main.c
1 /*
2  * Copyright 2008 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * This program is free software; you may redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; version 2 of the License.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
16  * SOFTWARE.
17  *
18  */
19
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/workqueue.h>
27 #include <linux/pci.h>
28 #include <linux/netdevice.h>
29 #include <linux/etherdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/if_vlan.h>
32 #include <linux/ethtool.h>
33 #include <linux/in.h>
34 #include <linux/ip.h>
35 #include <linux/ipv6.h>
36 #include <linux/tcp.h>
37 #include <net/ip6_checksum.h>
38
39 #include "cq_enet_desc.h"
40 #include "vnic_dev.h"
41 #include "vnic_intr.h"
42 #include "vnic_stats.h"
43 #include "enic_res.h"
44 #include "enic.h"
45
46 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
47
48 /* Supported devices */
49 static struct pci_device_id enic_id_table[] = {
50         { PCI_VDEVICE(CISCO, 0x0043) },
51         { 0, }  /* end of table */
52 };
53
54 MODULE_DESCRIPTION(DRV_DESCRIPTION);
55 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
56 MODULE_LICENSE("GPL");
57 MODULE_VERSION(DRV_VERSION);
58 MODULE_DEVICE_TABLE(pci, enic_id_table);
59
60 struct enic_stat {
61         char name[ETH_GSTRING_LEN];
62         unsigned int offset;
63 };
64
65 #define ENIC_TX_STAT(stat)      \
66         { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
67 #define ENIC_RX_STAT(stat)      \
68         { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
69
70 static const struct enic_stat enic_tx_stats[] = {
71         ENIC_TX_STAT(tx_frames_ok),
72         ENIC_TX_STAT(tx_unicast_frames_ok),
73         ENIC_TX_STAT(tx_multicast_frames_ok),
74         ENIC_TX_STAT(tx_broadcast_frames_ok),
75         ENIC_TX_STAT(tx_bytes_ok),
76         ENIC_TX_STAT(tx_unicast_bytes_ok),
77         ENIC_TX_STAT(tx_multicast_bytes_ok),
78         ENIC_TX_STAT(tx_broadcast_bytes_ok),
79         ENIC_TX_STAT(tx_drops),
80         ENIC_TX_STAT(tx_errors),
81         ENIC_TX_STAT(tx_tso),
82 };
83
84 static const struct enic_stat enic_rx_stats[] = {
85         ENIC_RX_STAT(rx_frames_ok),
86         ENIC_RX_STAT(rx_frames_total),
87         ENIC_RX_STAT(rx_unicast_frames_ok),
88         ENIC_RX_STAT(rx_multicast_frames_ok),
89         ENIC_RX_STAT(rx_broadcast_frames_ok),
90         ENIC_RX_STAT(rx_bytes_ok),
91         ENIC_RX_STAT(rx_unicast_bytes_ok),
92         ENIC_RX_STAT(rx_multicast_bytes_ok),
93         ENIC_RX_STAT(rx_broadcast_bytes_ok),
94         ENIC_RX_STAT(rx_drop),
95         ENIC_RX_STAT(rx_no_bufs),
96         ENIC_RX_STAT(rx_errors),
97         ENIC_RX_STAT(rx_rss),
98         ENIC_RX_STAT(rx_crc_errors),
99         ENIC_RX_STAT(rx_frames_64),
100         ENIC_RX_STAT(rx_frames_127),
101         ENIC_RX_STAT(rx_frames_255),
102         ENIC_RX_STAT(rx_frames_511),
103         ENIC_RX_STAT(rx_frames_1023),
104         ENIC_RX_STAT(rx_frames_1518),
105         ENIC_RX_STAT(rx_frames_to_max),
106 };
107
108 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
109 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
110
111 static int enic_get_settings(struct net_device *netdev,
112         struct ethtool_cmd *ecmd)
113 {
114         struct enic *enic = netdev_priv(netdev);
115
116         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
117         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
118         ecmd->port = PORT_FIBRE;
119         ecmd->transceiver = XCVR_EXTERNAL;
120
121         if (netif_carrier_ok(netdev)) {
122                 ecmd->speed = vnic_dev_port_speed(enic->vdev);
123                 ecmd->duplex = DUPLEX_FULL;
124         } else {
125                 ecmd->speed = -1;
126                 ecmd->duplex = -1;
127         }
128
129         ecmd->autoneg = AUTONEG_DISABLE;
130
131         return 0;
132 }
133
134 static void enic_get_drvinfo(struct net_device *netdev,
135         struct ethtool_drvinfo *drvinfo)
136 {
137         struct enic *enic = netdev_priv(netdev);
138         struct vnic_devcmd_fw_info *fw_info;
139
140         spin_lock(&enic->devcmd_lock);
141         vnic_dev_fw_info(enic->vdev, &fw_info);
142         spin_unlock(&enic->devcmd_lock);
143
144         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
145         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
146         strncpy(drvinfo->fw_version, fw_info->fw_version,
147                 sizeof(drvinfo->fw_version));
148         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
149                 sizeof(drvinfo->bus_info));
150 }
151
152 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
153 {
154         unsigned int i;
155
156         switch (stringset) {
157         case ETH_SS_STATS:
158                 for (i = 0; i < enic_n_tx_stats; i++) {
159                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
160                         data += ETH_GSTRING_LEN;
161                 }
162                 for (i = 0; i < enic_n_rx_stats; i++) {
163                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
164                         data += ETH_GSTRING_LEN;
165                 }
166                 break;
167         }
168 }
169
170 static int enic_get_sset_count(struct net_device *netdev, int sset)
171 {
172         switch (sset) {
173         case ETH_SS_STATS:
174                 return enic_n_tx_stats + enic_n_rx_stats;
175         default:
176                 return -EOPNOTSUPP;
177         }
178 }
179
180 static void enic_get_ethtool_stats(struct net_device *netdev,
181         struct ethtool_stats *stats, u64 *data)
182 {
183         struct enic *enic = netdev_priv(netdev);
184         struct vnic_stats *vstats;
185         unsigned int i;
186
187         spin_lock(&enic->devcmd_lock);
188         vnic_dev_stats_dump(enic->vdev, &vstats);
189         spin_unlock(&enic->devcmd_lock);
190
191         for (i = 0; i < enic_n_tx_stats; i++)
192                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
193         for (i = 0; i < enic_n_rx_stats; i++)
194                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
195 }
196
197 static u32 enic_get_rx_csum(struct net_device *netdev)
198 {
199         struct enic *enic = netdev_priv(netdev);
200         return enic->csum_rx_enabled;
201 }
202
203 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
204 {
205         struct enic *enic = netdev_priv(netdev);
206
207         if (data && !ENIC_SETTING(enic, RXCSUM))
208                 return -EINVAL;
209
210         enic->csum_rx_enabled = !!data;
211
212         return 0;
213 }
214
215 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
216 {
217         struct enic *enic = netdev_priv(netdev);
218
219         if (data && !ENIC_SETTING(enic, TXCSUM))
220                 return -EINVAL;
221
222         if (data)
223                 netdev->features |= NETIF_F_HW_CSUM;
224         else
225                 netdev->features &= ~NETIF_F_HW_CSUM;
226
227         return 0;
228 }
229
230 static int enic_set_tso(struct net_device *netdev, u32 data)
231 {
232         struct enic *enic = netdev_priv(netdev);
233
234         if (data && !ENIC_SETTING(enic, TSO))
235                 return -EINVAL;
236
237         if (data)
238                 netdev->features |=
239                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
240         else
241                 netdev->features &=
242                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
243
244         return 0;
245 }
246
247 static u32 enic_get_msglevel(struct net_device *netdev)
248 {
249         struct enic *enic = netdev_priv(netdev);
250         return enic->msg_enable;
251 }
252
253 static void enic_set_msglevel(struct net_device *netdev, u32 value)
254 {
255         struct enic *enic = netdev_priv(netdev);
256         enic->msg_enable = value;
257 }
258
259 static struct ethtool_ops enic_ethtool_ops = {
260         .get_settings = enic_get_settings,
261         .get_drvinfo = enic_get_drvinfo,
262         .get_msglevel = enic_get_msglevel,
263         .set_msglevel = enic_set_msglevel,
264         .get_link = ethtool_op_get_link,
265         .get_strings = enic_get_strings,
266         .get_sset_count = enic_get_sset_count,
267         .get_ethtool_stats = enic_get_ethtool_stats,
268         .get_rx_csum = enic_get_rx_csum,
269         .set_rx_csum = enic_set_rx_csum,
270         .get_tx_csum = ethtool_op_get_tx_csum,
271         .set_tx_csum = enic_set_tx_csum,
272         .get_sg = ethtool_op_get_sg,
273         .set_sg = ethtool_op_set_sg,
274         .get_tso = ethtool_op_get_tso,
275         .set_tso = enic_set_tso,
276 };
277
278 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
279 {
280         struct enic *enic = vnic_dev_priv(wq->vdev);
281
282         if (buf->sop)
283                 pci_unmap_single(enic->pdev, buf->dma_addr,
284                         buf->len, PCI_DMA_TODEVICE);
285         else
286                 pci_unmap_page(enic->pdev, buf->dma_addr,
287                         buf->len, PCI_DMA_TODEVICE);
288
289         if (buf->os_buf)
290                 dev_kfree_skb_any(buf->os_buf);
291 }
292
293 static void enic_wq_free_buf(struct vnic_wq *wq,
294         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
295 {
296         enic_free_wq_buf(wq, buf);
297 }
298
299 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
300         u8 type, u16 q_number, u16 completed_index, void *opaque)
301 {
302         struct enic *enic = vnic_dev_priv(vdev);
303
304         spin_lock(&enic->wq_lock[q_number]);
305
306         vnic_wq_service(&enic->wq[q_number], cq_desc,
307                 completed_index, enic_wq_free_buf,
308                 opaque);
309
310         if (netif_queue_stopped(enic->netdev) &&
311             vnic_wq_desc_avail(&enic->wq[q_number]) >= MAX_SKB_FRAGS + 1)
312                 netif_wake_queue(enic->netdev);
313
314         spin_unlock(&enic->wq_lock[q_number]);
315
316         return 0;
317 }
318
319 static void enic_log_q_error(struct enic *enic)
320 {
321         unsigned int i;
322         u32 error_status;
323
324         for (i = 0; i < enic->wq_count; i++) {
325                 error_status = vnic_wq_error_status(&enic->wq[i]);
326                 if (error_status)
327                         printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
328                                 enic->netdev->name, i, error_status);
329         }
330
331         for (i = 0; i < enic->rq_count; i++) {
332                 error_status = vnic_rq_error_status(&enic->rq[i]);
333                 if (error_status)
334                         printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
335                                 enic->netdev->name, i, error_status);
336         }
337 }
338
339 static void enic_link_check(struct enic *enic)
340 {
341         int link_status = vnic_dev_link_status(enic->vdev);
342         int carrier_ok = netif_carrier_ok(enic->netdev);
343
344         if (link_status && !carrier_ok) {
345                 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
346                 netif_carrier_on(enic->netdev);
347         } else if (!link_status && carrier_ok) {
348                 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
349                 netif_carrier_off(enic->netdev);
350         }
351 }
352
353 static void enic_mtu_check(struct enic *enic)
354 {
355         u32 mtu = vnic_dev_mtu(enic->vdev);
356
357         if (mtu != enic->port_mtu) {
358                 if (mtu < enic->netdev->mtu)
359                         printk(KERN_WARNING PFX
360                                 "%s: interface MTU (%d) set higher "
361                                 "than switch port MTU (%d)\n",
362                                 enic->netdev->name, enic->netdev->mtu, mtu);
363                 enic->port_mtu = mtu;
364         }
365 }
366
367 static void enic_msglvl_check(struct enic *enic)
368 {
369         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
370
371         if (msg_enable != enic->msg_enable) {
372                 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
373                         enic->netdev->name, enic->msg_enable, msg_enable);
374                 enic->msg_enable = msg_enable;
375         }
376 }
377
378 static void enic_notify_check(struct enic *enic)
379 {
380         enic_msglvl_check(enic);
381         enic_mtu_check(enic);
382         enic_link_check(enic);
383 }
384
385 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
386
387 static irqreturn_t enic_isr_legacy(int irq, void *data)
388 {
389         struct net_device *netdev = data;
390         struct enic *enic = netdev_priv(netdev);
391         u32 pba;
392
393         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
394
395         pba = vnic_intr_legacy_pba(enic->legacy_pba);
396         if (!pba) {
397                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
398                 return IRQ_NONE;        /* not our interrupt */
399         }
400
401         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY))
402                 enic_notify_check(enic);
403
404         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
405                 enic_log_q_error(enic);
406                 /* schedule recovery from WQ/RQ error */
407                 schedule_work(&enic->reset);
408                 return IRQ_HANDLED;
409         }
410
411         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
412                 if (netif_rx_schedule_prep(netdev, &enic->napi))
413                         __netif_rx_schedule(netdev, &enic->napi);
414         } else {
415                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
416         }
417
418         return IRQ_HANDLED;
419 }
420
421 static irqreturn_t enic_isr_msi(int irq, void *data)
422 {
423         struct enic *enic = data;
424
425         /* With MSI, there is no sharing of interrupts, so this is
426          * our interrupt and there is no need to ack it.  The device
427          * is not providing per-vector masking, so the OS will not
428          * write to PCI config space to mask/unmask the interrupt.
429          * We're using mask_on_assertion for MSI, so the device
430          * automatically masks the interrupt when the interrupt is
431          * generated.  Later, when exiting polling, the interrupt
432          * will be unmasked (see enic_poll).
433          *
434          * Also, the device uses the same PCIe Traffic Class (TC)
435          * for Memory Write data and MSI, so there are no ordering
436          * issues; the MSI will always arrive at the Root Complex
437          * _after_ corresponding Memory Writes (i.e. descriptor
438          * writes).
439          */
440
441         netif_rx_schedule(enic->netdev, &enic->napi);
442
443         return IRQ_HANDLED;
444 }
445
446 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
447 {
448         struct enic *enic = data;
449
450         /* schedule NAPI polling for RQ cleanup */
451         netif_rx_schedule(enic->netdev, &enic->napi);
452
453         return IRQ_HANDLED;
454 }
455
456 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
457 {
458         struct enic *enic = data;
459         unsigned int wq_work_to_do = -1; /* no limit */
460         unsigned int wq_work_done;
461
462         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
463                 wq_work_to_do, enic_wq_service, NULL);
464
465         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
466                 wq_work_done,
467                 1 /* unmask intr */,
468                 1 /* reset intr timer */);
469
470         return IRQ_HANDLED;
471 }
472
473 static irqreturn_t enic_isr_msix_err(int irq, void *data)
474 {
475         struct enic *enic = data;
476
477         enic_log_q_error(enic);
478
479         /* schedule recovery from WQ/RQ error */
480         schedule_work(&enic->reset);
481
482         return IRQ_HANDLED;
483 }
484
485 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
486 {
487         struct enic *enic = data;
488
489         enic_notify_check(enic);
490         vnic_intr_unmask(&enic->intr[ENIC_MSIX_NOTIFY]);
491
492         return IRQ_HANDLED;
493 }
494
495 static inline void enic_queue_wq_skb_cont(struct enic *enic,
496         struct vnic_wq *wq, struct sk_buff *skb,
497         unsigned int len_left)
498 {
499         skb_frag_t *frag;
500
501         /* Queue additional data fragments */
502         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
503                 len_left -= frag->size;
504                 enic_queue_wq_desc_cont(wq, skb,
505                         pci_map_page(enic->pdev, frag->page,
506                                 frag->page_offset, frag->size,
507                                 PCI_DMA_TODEVICE),
508                         frag->size,
509                         (len_left == 0));       /* EOP? */
510         }
511 }
512
513 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
514         struct vnic_wq *wq, struct sk_buff *skb,
515         int vlan_tag_insert, unsigned int vlan_tag)
516 {
517         unsigned int head_len = skb_headlen(skb);
518         unsigned int len_left = skb->len - head_len;
519         int eop = (len_left == 0);
520
521         /* Queue the main skb fragment */
522         enic_queue_wq_desc(wq, skb,
523                 pci_map_single(enic->pdev, skb->data,
524                         head_len, PCI_DMA_TODEVICE),
525                 head_len,
526                 vlan_tag_insert, vlan_tag,
527                 eop);
528
529         if (!eop)
530                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
531 }
532
533 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
534         struct vnic_wq *wq, struct sk_buff *skb,
535         int vlan_tag_insert, unsigned int vlan_tag)
536 {
537         unsigned int head_len = skb_headlen(skb);
538         unsigned int len_left = skb->len - head_len;
539         unsigned int hdr_len = skb_transport_offset(skb);
540         unsigned int csum_offset = hdr_len + skb->csum_offset;
541         int eop = (len_left == 0);
542
543         /* Queue the main skb fragment */
544         enic_queue_wq_desc_csum_l4(wq, skb,
545                 pci_map_single(enic->pdev, skb->data,
546                         head_len, PCI_DMA_TODEVICE),
547                 head_len,
548                 csum_offset,
549                 hdr_len,
550                 vlan_tag_insert, vlan_tag,
551                 eop);
552
553         if (!eop)
554                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
555 }
556
557 static inline void enic_queue_wq_skb_tso(struct enic *enic,
558         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
559         int vlan_tag_insert, unsigned int vlan_tag)
560 {
561         unsigned int head_len = skb_headlen(skb);
562         unsigned int len_left = skb->len - head_len;
563         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
564         int eop = (len_left == 0);
565
566         /* Preload TCP csum field with IP pseudo hdr calculated
567          * with IP length set to zero.  HW will later add in length
568          * to each TCP segment resulting from the TSO.
569          */
570
571         if (skb->protocol == __constant_htons(ETH_P_IP)) {
572                 ip_hdr(skb)->check = 0;
573                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
574                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
575         } else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
576                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
577                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
578         }
579
580         /* Queue the main skb fragment */
581         enic_queue_wq_desc_tso(wq, skb,
582                 pci_map_single(enic->pdev, skb->data,
583                         head_len, PCI_DMA_TODEVICE),
584                 head_len,
585                 mss, hdr_len,
586                 vlan_tag_insert, vlan_tag,
587                 eop);
588
589         if (!eop)
590                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
591 }
592
593 static inline void enic_queue_wq_skb(struct enic *enic,
594         struct vnic_wq *wq, struct sk_buff *skb)
595 {
596         unsigned int mss = skb_shinfo(skb)->gso_size;
597         unsigned int vlan_tag = 0;
598         int vlan_tag_insert = 0;
599
600         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
601                 /* VLAN tag from trunking driver */
602                 vlan_tag_insert = 1;
603                 vlan_tag = vlan_tx_tag_get(skb);
604         }
605
606         if (mss)
607                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
608                         vlan_tag_insert, vlan_tag);
609         else if (skb->ip_summed == CHECKSUM_PARTIAL)
610                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
611                         vlan_tag_insert, vlan_tag);
612         else
613                 enic_queue_wq_skb_vlan(enic, wq, skb,
614                         vlan_tag_insert, vlan_tag);
615 }
616
617 /* netif_tx_lock held, process context with BHs disabled */
618 static int enic_hard_start_xmit(struct sk_buff *skb, struct net_device *netdev)
619 {
620         struct enic *enic = netdev_priv(netdev);
621         struct vnic_wq *wq = &enic->wq[0];
622         unsigned long flags;
623
624         if (skb->len <= 0) {
625                 dev_kfree_skb(skb);
626                 return NETDEV_TX_OK;
627         }
628
629         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
630          * which is very likely.  In the off chance it's going to take
631          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
632          */
633
634         if (skb_shinfo(skb)->gso_size == 0 &&
635             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
636             skb_linearize(skb)) {
637                 dev_kfree_skb(skb);
638                 return NETDEV_TX_OK;
639         }
640
641         spin_lock_irqsave(&enic->wq_lock[0], flags);
642
643         if (vnic_wq_desc_avail(wq) < skb_shinfo(skb)->nr_frags + 1) {
644                 netif_stop_queue(netdev);
645                 /* This is a hard error, log it */
646                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
647                         "queue awake!\n", netdev->name);
648                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
649                 return NETDEV_TX_BUSY;
650         }
651
652         enic_queue_wq_skb(enic, wq, skb);
653
654         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + 1)
655                 netif_stop_queue(netdev);
656
657         netdev->trans_start = jiffies;
658
659         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
660
661         return NETDEV_TX_OK;
662 }
663
664 /* dev_base_lock rwlock held, nominally process context */
665 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
666 {
667         struct enic *enic = netdev_priv(netdev);
668         struct net_device_stats *net_stats = &netdev->stats;
669         struct vnic_stats *stats;
670
671         spin_lock(&enic->devcmd_lock);
672         vnic_dev_stats_dump(enic->vdev, &stats);
673         spin_unlock(&enic->devcmd_lock);
674
675         net_stats->tx_packets = stats->tx.tx_frames_ok;
676         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
677         net_stats->tx_errors = stats->tx.tx_errors;
678         net_stats->tx_dropped = stats->tx.tx_drops;
679
680         net_stats->rx_packets = stats->rx.rx_frames_ok;
681         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
682         net_stats->rx_errors = stats->rx.rx_errors;
683         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
684         net_stats->rx_crc_errors = stats->rx.rx_crc_errors;
685         net_stats->rx_dropped = stats->rx.rx_no_bufs;
686
687         return net_stats;
688 }
689
690 static void enic_reset_mcaddrs(struct enic *enic)
691 {
692         enic->mc_count = 0;
693 }
694
695 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
696 {
697         if (!is_valid_ether_addr(addr))
698                 return -EADDRNOTAVAIL;
699
700         memcpy(netdev->dev_addr, addr, netdev->addr_len);
701
702         return 0;
703 }
704
705 /* netif_tx_lock held, BHs disabled */
706 static void enic_set_multicast_list(struct net_device *netdev)
707 {
708         struct enic *enic = netdev_priv(netdev);
709         struct dev_mc_list *list = netdev->mc_list;
710         int directed = 1;
711         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
712         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
713         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
714         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
715             (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
716         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
717         unsigned int mc_count = netdev->mc_count;
718         unsigned int i, j;
719
720         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
721                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
722
723         spin_lock(&enic->devcmd_lock);
724
725         vnic_dev_packet_filter(enic->vdev, directed,
726                 multicast, broadcast, promisc, allmulti);
727
728         /* Is there an easier way?  Trying to minimize to
729          * calls to add/del multicast addrs.  We keep the
730          * addrs from the last call in enic->mc_addr and
731          * look for changes to add/del.
732          */
733
734         for (i = 0; list && i < mc_count; i++) {
735                 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
736                 list = list->next;
737         }
738
739         for (i = 0; i < enic->mc_count; i++) {
740                 for (j = 0; j < mc_count; j++)
741                         if (compare_ether_addr(enic->mc_addr[i],
742                                 mc_addr[j]) == 0)
743                                 break;
744                 if (j == mc_count)
745                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
746         }
747
748         for (i = 0; i < mc_count; i++) {
749                 for (j = 0; j < enic->mc_count; j++)
750                         if (compare_ether_addr(mc_addr[i],
751                                 enic->mc_addr[j]) == 0)
752                                 break;
753                 if (j == enic->mc_count)
754                         enic_add_multicast_addr(enic, mc_addr[i]);
755         }
756
757         /* Save the list to compare against next time
758          */
759
760         for (i = 0; i < mc_count; i++)
761                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
762
763         enic->mc_count = mc_count;
764
765         spin_unlock(&enic->devcmd_lock);
766 }
767
768 /* rtnl lock is held */
769 static void enic_vlan_rx_register(struct net_device *netdev,
770         struct vlan_group *vlan_group)
771 {
772         struct enic *enic = netdev_priv(netdev);
773         enic->vlan_group = vlan_group;
774 }
775
776 /* rtnl lock is held */
777 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
778 {
779         struct enic *enic = netdev_priv(netdev);
780
781         spin_lock(&enic->devcmd_lock);
782         enic_add_vlan(enic, vid);
783         spin_unlock(&enic->devcmd_lock);
784 }
785
786 /* rtnl lock is held */
787 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
788 {
789         struct enic *enic = netdev_priv(netdev);
790
791         spin_lock(&enic->devcmd_lock);
792         enic_del_vlan(enic, vid);
793         spin_unlock(&enic->devcmd_lock);
794 }
795
796 /* netif_tx_lock held, BHs disabled */
797 static void enic_tx_timeout(struct net_device *netdev)
798 {
799         struct enic *enic = netdev_priv(netdev);
800         schedule_work(&enic->reset);
801 }
802
803 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
804 {
805         struct enic *enic = vnic_dev_priv(rq->vdev);
806
807         if (!buf->os_buf)
808                 return;
809
810         pci_unmap_single(enic->pdev, buf->dma_addr,
811                 buf->len, PCI_DMA_FROMDEVICE);
812         dev_kfree_skb_any(buf->os_buf);
813 }
814
815 static inline struct sk_buff *enic_rq_alloc_skb(unsigned int size)
816 {
817         struct sk_buff *skb;
818
819         skb = dev_alloc_skb(size + NET_IP_ALIGN);
820
821         if (skb)
822                 skb_reserve(skb, NET_IP_ALIGN);
823
824         return skb;
825 }
826
827 static int enic_rq_alloc_buf(struct vnic_rq *rq)
828 {
829         struct enic *enic = vnic_dev_priv(rq->vdev);
830         struct sk_buff *skb;
831         unsigned int len = enic->netdev->mtu + ETH_HLEN;
832         unsigned int os_buf_index = 0;
833         dma_addr_t dma_addr;
834
835         skb = enic_rq_alloc_skb(len);
836         if (!skb)
837                 return -ENOMEM;
838
839         dma_addr = pci_map_single(enic->pdev, skb->data,
840                 len, PCI_DMA_FROMDEVICE);
841
842         enic_queue_rq_desc(rq, skb, os_buf_index,
843                 dma_addr, len);
844
845         return 0;
846 }
847
848 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
849         void **tcph, u64 *hdr_flags, void *priv)
850 {
851         struct cq_enet_rq_desc *cq_desc = priv;
852         unsigned int ip_len;
853         struct iphdr *iph;
854
855         u8 type, color, eop, sop, ingress_port, vlan_stripped;
856         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
857         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
858         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
859         u8 packet_error;
860         u16 q_number, completed_index, bytes_written, vlan, checksum;
861         u32 rss_hash;
862
863         cq_enet_rq_desc_dec(cq_desc,
864                 &type, &color, &q_number, &completed_index,
865                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
866                 &csum_not_calc, &rss_hash, &bytes_written,
867                 &packet_error, &vlan_stripped, &vlan, &checksum,
868                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
869                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
870                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
871                 &fcs_ok);
872
873         if (!(ipv4 && tcp && !ipv4_fragment))
874                 return -1;
875
876         skb_reset_network_header(skb);
877         iph = ip_hdr(skb);
878
879         ip_len = ip_hdrlen(skb);
880         skb_set_transport_header(skb, ip_len);
881
882         /* check if ip header and tcp header are complete */
883         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
884                 return -1;
885
886         *hdr_flags = LRO_IPV4 | LRO_TCP;
887         *tcph = tcp_hdr(skb);
888         *iphdr = iph;
889
890         return 0;
891 }
892
893 static void enic_rq_indicate_buf(struct vnic_rq *rq,
894         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
895         int skipped, void *opaque)
896 {
897         struct enic *enic = vnic_dev_priv(rq->vdev);
898         struct sk_buff *skb;
899
900         u8 type, color, eop, sop, ingress_port, vlan_stripped;
901         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
902         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
903         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
904         u8 packet_error;
905         u16 q_number, completed_index, bytes_written, vlan, checksum;
906         u32 rss_hash;
907
908         if (skipped)
909                 return;
910
911         skb = buf->os_buf;
912         prefetch(skb->data - NET_IP_ALIGN);
913         pci_unmap_single(enic->pdev, buf->dma_addr,
914                 buf->len, PCI_DMA_FROMDEVICE);
915
916         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
917                 &type, &color, &q_number, &completed_index,
918                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
919                 &csum_not_calc, &rss_hash, &bytes_written,
920                 &packet_error, &vlan_stripped, &vlan, &checksum,
921                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
922                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
923                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
924                 &fcs_ok);
925
926         if (packet_error) {
927
928                 if (bytes_written > 0 && !fcs_ok) {
929                         if (net_ratelimit())
930                                 printk(KERN_ERR PFX
931                                         "%s: packet error: bad FCS\n",
932                                         enic->netdev->name);
933                 }
934
935                 dev_kfree_skb_any(skb);
936
937                 return;
938         }
939
940         if (eop && bytes_written > 0) {
941
942                 /* Good receive
943                  */
944
945                 skb_put(skb, bytes_written);
946                 skb->protocol = eth_type_trans(skb, enic->netdev);
947
948                 if (enic->csum_rx_enabled && !csum_not_calc) {
949                         skb->csum = htons(checksum);
950                         skb->ip_summed = CHECKSUM_COMPLETE;
951                 }
952
953                 skb->dev = enic->netdev;
954                 enic->netdev->last_rx = jiffies;
955
956                 if (enic->vlan_group && vlan_stripped) {
957
958                         if (ENIC_SETTING(enic, LRO) && ipv4)
959                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
960                                         skb, enic->vlan_group,
961                                         vlan, cq_desc);
962                         else
963                                 vlan_hwaccel_receive_skb(skb,
964                                         enic->vlan_group, vlan);
965
966                 } else {
967
968                         if (ENIC_SETTING(enic, LRO) && ipv4)
969                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
970                         else
971                                 netif_receive_skb(skb);
972
973                 }
974
975         } else {
976
977                 /* Buffer overflow
978                  */
979
980                 dev_kfree_skb_any(skb);
981         }
982 }
983
984 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
985         u8 type, u16 q_number, u16 completed_index, void *opaque)
986 {
987         struct enic *enic = vnic_dev_priv(vdev);
988
989         vnic_rq_service(&enic->rq[q_number], cq_desc,
990                 completed_index, VNIC_RQ_RETURN_DESC,
991                 enic_rq_indicate_buf, opaque);
992
993         return 0;
994 }
995
996 static void enic_rq_drop_buf(struct vnic_rq *rq,
997         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
998         int skipped, void *opaque)
999 {
1000         struct enic *enic = vnic_dev_priv(rq->vdev);
1001         struct sk_buff *skb = buf->os_buf;
1002
1003         if (skipped)
1004                 return;
1005
1006         pci_unmap_single(enic->pdev, buf->dma_addr,
1007                 buf->len, PCI_DMA_FROMDEVICE);
1008
1009         dev_kfree_skb_any(skb);
1010 }
1011
1012 static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1013         u8 type, u16 q_number, u16 completed_index, void *opaque)
1014 {
1015         struct enic *enic = vnic_dev_priv(vdev);
1016
1017         vnic_rq_service(&enic->rq[q_number], cq_desc,
1018                 completed_index, VNIC_RQ_RETURN_DESC,
1019                 enic_rq_drop_buf, opaque);
1020
1021         return 0;
1022 }
1023
1024 static int enic_poll(struct napi_struct *napi, int budget)
1025 {
1026         struct enic *enic = container_of(napi, struct enic, napi);
1027         struct net_device *netdev = enic->netdev;
1028         unsigned int rq_work_to_do = budget;
1029         unsigned int wq_work_to_do = -1; /* no limit */
1030         unsigned int  work_done, rq_work_done, wq_work_done;
1031
1032         /* Service RQ (first) and WQ
1033          */
1034
1035         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1036                 rq_work_to_do, enic_rq_service, NULL);
1037
1038         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1039                 wq_work_to_do, enic_wq_service, NULL);
1040
1041         /* Accumulate intr event credits for this polling
1042          * cycle.  An intr event is the completion of a
1043          * a WQ or RQ packet.
1044          */
1045
1046         work_done = rq_work_done + wq_work_done;
1047
1048         if (work_done > 0)
1049                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1050                         work_done,
1051                         0 /* don't unmask intr */,
1052                         0 /* don't reset intr timer */);
1053
1054         if (rq_work_done > 0) {
1055
1056                 /* Replenish RQ
1057                  */
1058
1059                 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1060
1061         } else {
1062
1063                 /* If no work done, flush all LROs and exit polling
1064                  */
1065
1066                 if (ENIC_SETTING(enic, LRO))
1067                         lro_flush_all(&enic->lro_mgr);
1068
1069                 netif_rx_complete(netdev, napi);
1070                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1071         }
1072
1073         return rq_work_done;
1074 }
1075
1076 static int enic_poll_msix(struct napi_struct *napi, int budget)
1077 {
1078         struct enic *enic = container_of(napi, struct enic, napi);
1079         struct net_device *netdev = enic->netdev;
1080         unsigned int work_to_do = budget;
1081         unsigned int work_done;
1082
1083         /* Service RQ
1084          */
1085
1086         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1087                 work_to_do, enic_rq_service, NULL);
1088
1089         if (work_done > 0) {
1090
1091                 /* Replenish RQ
1092                  */
1093
1094                 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1095
1096                 /* Accumulate intr event credits for this polling
1097                  * cycle.  An intr event is the completion of a
1098                  * a WQ or RQ packet.
1099                  */
1100
1101                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1102                         work_done,
1103                         0 /* don't unmask intr */,
1104                         0 /* don't reset intr timer */);
1105         } else {
1106
1107                 /* If no work done, flush all LROs and exit polling
1108                  */
1109
1110                 if (ENIC_SETTING(enic, LRO))
1111                         lro_flush_all(&enic->lro_mgr);
1112
1113                 netif_rx_complete(netdev, napi);
1114                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1115         }
1116
1117         return work_done;
1118 }
1119
1120 static void enic_notify_timer(unsigned long data)
1121 {
1122         struct enic *enic = (struct enic *)data;
1123
1124         enic_notify_check(enic);
1125
1126         mod_timer(&enic->notify_timer,
1127                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1128 }
1129
1130 static void enic_free_intr(struct enic *enic)
1131 {
1132         struct net_device *netdev = enic->netdev;
1133         unsigned int i;
1134
1135         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1136         case VNIC_DEV_INTR_MODE_INTX:
1137                 free_irq(enic->pdev->irq, netdev);
1138                 break;
1139         case VNIC_DEV_INTR_MODE_MSI:
1140                 free_irq(enic->pdev->irq, enic);
1141                 break;
1142         case VNIC_DEV_INTR_MODE_MSIX:
1143                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1144                         if (enic->msix[i].requested)
1145                                 free_irq(enic->msix_entry[i].vector,
1146                                         enic->msix[i].devid);
1147                 break;
1148         default:
1149                 break;
1150         }
1151 }
1152
1153 static int enic_request_intr(struct enic *enic)
1154 {
1155         struct net_device *netdev = enic->netdev;
1156         unsigned int i;
1157         int err = 0;
1158
1159         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1160
1161         case VNIC_DEV_INTR_MODE_INTX:
1162
1163                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1164                         IRQF_SHARED, netdev->name, netdev);
1165                 break;
1166
1167         case VNIC_DEV_INTR_MODE_MSI:
1168
1169                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1170                         0, netdev->name, enic);
1171                 break;
1172
1173         case VNIC_DEV_INTR_MODE_MSIX:
1174
1175                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1176                         "%.11s-rx-0", netdev->name);
1177                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1178                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1179
1180                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1181                         "%.11s-tx-0", netdev->name);
1182                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1183                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1184
1185                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1186                         "%.11s-err", netdev->name);
1187                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1188                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1189
1190                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1191                         "%.11s-notify", netdev->name);
1192                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1193                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1194
1195                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1196                         err = request_irq(enic->msix_entry[i].vector,
1197                                 enic->msix[i].isr, 0,
1198                                 enic->msix[i].devname,
1199                                 enic->msix[i].devid);
1200                         if (err) {
1201                                 enic_free_intr(enic);
1202                                 break;
1203                         }
1204                         enic->msix[i].requested = 1;
1205                 }
1206
1207                 break;
1208
1209         default:
1210                 break;
1211         }
1212
1213         return err;
1214 }
1215
1216 static int enic_notify_set(struct enic *enic)
1217 {
1218         int err;
1219
1220         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1221         case VNIC_DEV_INTR_MODE_INTX:
1222                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1223                 break;
1224         case VNIC_DEV_INTR_MODE_MSIX:
1225                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1226                 break;
1227         default:
1228                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1229                 break;
1230         }
1231
1232         return err;
1233 }
1234
1235 static void enic_notify_timer_start(struct enic *enic)
1236 {
1237         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1238         case VNIC_DEV_INTR_MODE_MSI:
1239                 mod_timer(&enic->notify_timer, jiffies);
1240                 break;
1241         default:
1242                 /* Using intr for notification for INTx/MSI-X */
1243                 break;
1244         };
1245 }
1246
1247 /* rtnl lock is held, process context */
1248 static int enic_open(struct net_device *netdev)
1249 {
1250         struct enic *enic = netdev_priv(netdev);
1251         unsigned int i;
1252         int err;
1253
1254         err = enic_request_intr(enic);
1255         if (err) {
1256                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1257                         netdev->name);
1258                 return err;
1259         }
1260
1261         err = enic_notify_set(enic);
1262         if (err) {
1263                 printk(KERN_ERR PFX
1264                         "%s: Failed to alloc notify buffer, aborting.\n",
1265                         netdev->name);
1266                 goto err_out_free_intr;
1267         }
1268
1269         for (i = 0; i < enic->rq_count; i++) {
1270                 err = vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1271                 if (err) {
1272                         printk(KERN_ERR PFX
1273                                 "%s: Unable to alloc receive buffers.\n",
1274                                 netdev->name);
1275                         goto err_out_notify_unset;
1276                 }
1277         }
1278
1279         for (i = 0; i < enic->wq_count; i++)
1280                 vnic_wq_enable(&enic->wq[i]);
1281         for (i = 0; i < enic->rq_count; i++)
1282                 vnic_rq_enable(&enic->rq[i]);
1283
1284         enic_add_station_addr(enic);
1285         enic_set_multicast_list(netdev);
1286
1287         netif_wake_queue(netdev);
1288         napi_enable(&enic->napi);
1289         vnic_dev_enable(enic->vdev);
1290
1291         for (i = 0; i < enic->intr_count; i++)
1292                 vnic_intr_unmask(&enic->intr[i]);
1293
1294         enic_notify_timer_start(enic);
1295
1296         return 0;
1297
1298 err_out_notify_unset:
1299         vnic_dev_notify_unset(enic->vdev);
1300 err_out_free_intr:
1301         enic_free_intr(enic);
1302
1303         return err;
1304 }
1305
1306 /* rtnl lock is held, process context */
1307 static int enic_stop(struct net_device *netdev)
1308 {
1309         struct enic *enic = netdev_priv(netdev);
1310         unsigned int i;
1311         int err;
1312
1313         del_timer_sync(&enic->notify_timer);
1314
1315         vnic_dev_disable(enic->vdev);
1316         napi_disable(&enic->napi);
1317         netif_stop_queue(netdev);
1318
1319         for (i = 0; i < enic->intr_count; i++)
1320                 vnic_intr_mask(&enic->intr[i]);
1321
1322         for (i = 0; i < enic->wq_count; i++) {
1323                 err = vnic_wq_disable(&enic->wq[i]);
1324                 if (err)
1325                         return err;
1326         }
1327         for (i = 0; i < enic->rq_count; i++) {
1328                 err = vnic_rq_disable(&enic->rq[i]);
1329                 if (err)
1330                         return err;
1331         }
1332
1333         vnic_dev_notify_unset(enic->vdev);
1334         enic_free_intr(enic);
1335
1336         (void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1337                 -1, enic_rq_service_drop, NULL);
1338         (void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1339                 -1, enic_wq_service, NULL);
1340
1341         for (i = 0; i < enic->wq_count; i++)
1342                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1343         for (i = 0; i < enic->rq_count; i++)
1344                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1345         for (i = 0; i < enic->cq_count; i++)
1346                 vnic_cq_clean(&enic->cq[i]);
1347         for (i = 0; i < enic->intr_count; i++)
1348                 vnic_intr_clean(&enic->intr[i]);
1349
1350         return 0;
1351 }
1352
1353 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1354 {
1355         struct enic *enic = netdev_priv(netdev);
1356         int running = netif_running(netdev);
1357
1358         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1359                 return -EINVAL;
1360
1361         if (running)
1362                 enic_stop(netdev);
1363
1364         netdev->mtu = new_mtu;
1365
1366         if (netdev->mtu > enic->port_mtu)
1367                 printk(KERN_WARNING PFX
1368                         "%s: interface MTU (%d) set higher "
1369                         "than port MTU (%d)\n",
1370                         netdev->name, netdev->mtu, enic->port_mtu);
1371
1372         if (running)
1373                 enic_open(netdev);
1374
1375         return 0;
1376 }
1377
1378 #ifdef CONFIG_NET_POLL_CONTROLLER
1379 static void enic_poll_controller(struct net_device *netdev)
1380 {
1381         struct enic *enic = netdev_priv(netdev);
1382         struct vnic_dev *vdev = enic->vdev;
1383
1384         switch (vnic_dev_get_intr_mode(vdev)) {
1385         case VNIC_DEV_INTR_MODE_MSIX:
1386                 enic_isr_msix_rq(enic->pdev->irq, enic);
1387                 enic_isr_msix_wq(enic->pdev->irq, enic);
1388                 break;
1389         case VNIC_DEV_INTR_MODE_MSI:
1390                 enic_isr_msi(enic->pdev->irq, enic);
1391                 break;
1392         case VNIC_DEV_INTR_MODE_INTX:
1393                 enic_isr_legacy(enic->pdev->irq, netdev);
1394                 break;
1395         default:
1396                 break;
1397         }
1398 }
1399 #endif
1400
1401 static int enic_dev_wait(struct vnic_dev *vdev,
1402         int (*start)(struct vnic_dev *, int),
1403         int (*finished)(struct vnic_dev *, int *),
1404         int arg)
1405 {
1406         unsigned long time;
1407         int done;
1408         int err;
1409
1410         BUG_ON(in_interrupt());
1411
1412         err = start(vdev, arg);
1413         if (err)
1414                 return err;
1415
1416         /* Wait for func to complete...2 seconds max
1417          */
1418
1419         time = jiffies + (HZ * 2);
1420         do {
1421
1422                 err = finished(vdev, &done);
1423                 if (err)
1424                         return err;
1425
1426                 if (done)
1427                         return 0;
1428
1429                 schedule_timeout_uninterruptible(HZ / 10);
1430
1431         } while (time_after(time, jiffies));
1432
1433         return -ETIMEDOUT;
1434 }
1435
1436 static int enic_dev_open(struct enic *enic)
1437 {
1438         int err;
1439
1440         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1441                 vnic_dev_open_done, 0);
1442         if (err)
1443                 printk(KERN_ERR PFX
1444                         "vNIC device open failed, err %d.\n", err);
1445
1446         return err;
1447 }
1448
1449 static int enic_dev_soft_reset(struct enic *enic)
1450 {
1451         int err;
1452
1453         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1454                 vnic_dev_soft_reset_done, 0);
1455         if (err)
1456                 printk(KERN_ERR PFX
1457                         "vNIC soft reset failed, err %d.\n", err);
1458
1459         return err;
1460 }
1461
1462 static void enic_reset(struct work_struct *work)
1463 {
1464         struct enic *enic = container_of(work, struct enic, reset);
1465
1466         if (!netif_running(enic->netdev))
1467                 return;
1468
1469         rtnl_lock();
1470
1471         spin_lock(&enic->devcmd_lock);
1472         vnic_dev_hang_notify(enic->vdev);
1473         spin_unlock(&enic->devcmd_lock);
1474
1475         enic_stop(enic->netdev);
1476         enic_dev_soft_reset(enic);
1477         enic_reset_mcaddrs(enic);
1478         enic_init_vnic_resources(enic);
1479         enic_open(enic->netdev);
1480
1481         rtnl_unlock();
1482 }
1483
1484 static int enic_set_intr_mode(struct enic *enic)
1485 {
1486         unsigned int n = ARRAY_SIZE(enic->rq);
1487         unsigned int m = ARRAY_SIZE(enic->wq);
1488         unsigned int i;
1489
1490         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1491          * system capabilities.
1492          *
1493          * Try MSI-X first
1494          *
1495          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1496          * (the second to last INTR is used for WQ/RQ errors)
1497          * (the last INTR is used for notifications)
1498          */
1499
1500         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1501         for (i = 0; i < n + m + 2; i++)
1502                 enic->msix_entry[i].entry = i;
1503
1504         if (enic->config.intr_mode < 1 &&
1505             enic->rq_count >= n &&
1506             enic->wq_count >= m &&
1507             enic->cq_count >= n + m &&
1508             enic->intr_count >= n + m + 2 &&
1509             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1510
1511                 enic->rq_count = n;
1512                 enic->wq_count = m;
1513                 enic->cq_count = n + m;
1514                 enic->intr_count = n + m + 2;
1515
1516                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1517
1518                 return 0;
1519         }
1520
1521         /* Next try MSI
1522          *
1523          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1524          */
1525
1526         if (enic->config.intr_mode < 2 &&
1527             enic->rq_count >= 1 &&
1528             enic->wq_count >= 1 &&
1529             enic->cq_count >= 2 &&
1530             enic->intr_count >= 1 &&
1531             !pci_enable_msi(enic->pdev)) {
1532
1533                 enic->rq_count = 1;
1534                 enic->wq_count = 1;
1535                 enic->cq_count = 2;
1536                 enic->intr_count = 1;
1537
1538                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1539
1540                 return 0;
1541         }
1542
1543         /* Next try INTx
1544          *
1545          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1546          * (the first INTR is used for WQ/RQ)
1547          * (the second INTR is used for WQ/RQ errors)
1548          * (the last INTR is used for notifications)
1549          */
1550
1551         if (enic->config.intr_mode < 3 &&
1552             enic->rq_count >= 1 &&
1553             enic->wq_count >= 1 &&
1554             enic->cq_count >= 2 &&
1555             enic->intr_count >= 3) {
1556
1557                 enic->rq_count = 1;
1558                 enic->wq_count = 1;
1559                 enic->cq_count = 2;
1560                 enic->intr_count = 3;
1561
1562                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1563
1564                 return 0;
1565         }
1566
1567         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1568
1569         return -EINVAL;
1570 }
1571
1572 static void enic_clear_intr_mode(struct enic *enic)
1573 {
1574         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1575         case VNIC_DEV_INTR_MODE_MSIX:
1576                 pci_disable_msix(enic->pdev);
1577                 break;
1578         case VNIC_DEV_INTR_MODE_MSI:
1579                 pci_disable_msi(enic->pdev);
1580                 break;
1581         default:
1582                 break;
1583         }
1584
1585         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1586 }
1587
1588 static void enic_iounmap(struct enic *enic)
1589 {
1590         if (enic->bar0.vaddr)
1591                 iounmap(enic->bar0.vaddr);
1592 }
1593
1594 static int __devinit enic_probe(struct pci_dev *pdev,
1595         const struct pci_device_id *ent)
1596 {
1597         struct net_device *netdev;
1598         struct enic *enic;
1599         int using_dac = 0;
1600         unsigned int i;
1601         int err;
1602
1603         const u8 rss_default_cpu = 0;
1604         const u8 rss_hash_type = 0;
1605         const u8 rss_hash_bits = 0;
1606         const u8 rss_base_cpu = 0;
1607         const u8 rss_enable = 0;
1608         const u8 tso_ipid_split_en = 0;
1609         const u8 ig_vlan_strip_en = 1;
1610
1611         /* Allocate net device structure and initialize.  Private
1612          * instance data is initialized to zero.
1613          */
1614
1615         netdev = alloc_etherdev(sizeof(struct enic));
1616         if (!netdev) {
1617                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1618                 return -ENOMEM;
1619         }
1620
1621         pci_set_drvdata(pdev, netdev);
1622
1623         SET_NETDEV_DEV(netdev, &pdev->dev);
1624
1625         enic = netdev_priv(netdev);
1626         enic->netdev = netdev;
1627         enic->pdev = pdev;
1628
1629         /* Setup PCI resources
1630          */
1631
1632         err = pci_enable_device(pdev);
1633         if (err) {
1634                 printk(KERN_ERR PFX
1635                         "Cannot enable PCI device, aborting.\n");
1636                 goto err_out_free_netdev;
1637         }
1638
1639         err = pci_request_regions(pdev, DRV_NAME);
1640         if (err) {
1641                 printk(KERN_ERR PFX
1642                         "Cannot request PCI regions, aborting.\n");
1643                 goto err_out_disable_device;
1644         }
1645
1646         pci_set_master(pdev);
1647
1648         /* Query PCI controller on system for DMA addressing
1649          * limitation for the device.  Try 40-bit first, and
1650          * fail to 32-bit.
1651          */
1652
1653         err = pci_set_dma_mask(pdev, DMA_40BIT_MASK);
1654         if (err) {
1655                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1656                 if (err) {
1657                         printk(KERN_ERR PFX
1658                                 "No usable DMA configuration, aborting.\n");
1659                         goto err_out_release_regions;
1660                 }
1661                 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1662                 if (err) {
1663                         printk(KERN_ERR PFX
1664                                 "Unable to obtain 32-bit DMA "
1665                                 "for consistent allocations, aborting.\n");
1666                         goto err_out_release_regions;
1667                 }
1668         } else {
1669                 err = pci_set_consistent_dma_mask(pdev, DMA_40BIT_MASK);
1670                 if (err) {
1671                         printk(KERN_ERR PFX
1672                                 "Unable to obtain 40-bit DMA "
1673                                 "for consistent allocations, aborting.\n");
1674                         goto err_out_release_regions;
1675                 }
1676                 using_dac = 1;
1677         }
1678
1679         /* Map vNIC resources from BAR0
1680          */
1681
1682         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1683                 printk(KERN_ERR PFX
1684                         "BAR0 not memory-map'able, aborting.\n");
1685                 err = -ENODEV;
1686                 goto err_out_release_regions;
1687         }
1688
1689         enic->bar0.vaddr = pci_iomap(pdev, 0, enic->bar0.len);
1690         enic->bar0.bus_addr = pci_resource_start(pdev, 0);
1691         enic->bar0.len = pci_resource_len(pdev, 0);
1692
1693         if (!enic->bar0.vaddr) {
1694                 printk(KERN_ERR PFX
1695                         "Cannot memory-map BAR0 res hdr, aborting.\n");
1696                 err = -ENODEV;
1697                 goto err_out_release_regions;
1698         }
1699
1700         /* Register vNIC device
1701          */
1702
1703         enic->vdev = vnic_dev_register(NULL, enic, pdev, &enic->bar0);
1704         if (!enic->vdev) {
1705                 printk(KERN_ERR PFX
1706                         "vNIC registration failed, aborting.\n");
1707                 err = -ENODEV;
1708                 goto err_out_iounmap;
1709         }
1710
1711         /* Issue device open to get device in known state
1712          */
1713
1714         err = enic_dev_open(enic);
1715         if (err) {
1716                 printk(KERN_ERR PFX
1717                         "vNIC dev open failed, aborting.\n");
1718                 goto err_out_vnic_unregister;
1719         }
1720
1721         /* Issue device init to initialize the vnic-to-switch link.
1722          * We'll start with carrier off and wait for link UP
1723          * notification later to turn on carrier.  We don't need
1724          * to wait here for the vnic-to-switch link initialization
1725          * to complete; link UP notification is the indication that
1726          * the process is complete.
1727          */
1728
1729         netif_carrier_off(netdev);
1730
1731         err = vnic_dev_init(enic->vdev, 0);
1732         if (err) {
1733                 printk(KERN_ERR PFX
1734                         "vNIC dev init failed, aborting.\n");
1735                 goto err_out_dev_close;
1736         }
1737
1738         /* Get vNIC configuration
1739          */
1740
1741         err = enic_get_vnic_config(enic);
1742         if (err) {
1743                 printk(KERN_ERR PFX
1744                         "Get vNIC configuration failed, aborting.\n");
1745                 goto err_out_dev_close;
1746         }
1747
1748         /* Get available resource counts
1749         */
1750
1751         enic_get_res_counts(enic);
1752
1753         /* Set interrupt mode based on resource counts and system
1754          * capabilities
1755         */
1756
1757         err = enic_set_intr_mode(enic);
1758         if (err) {
1759                 printk(KERN_ERR PFX
1760                         "Failed to set intr mode, aborting.\n");
1761                 goto err_out_dev_close;
1762         }
1763
1764         /* Allocate and configure vNIC resources
1765          */
1766
1767         err = enic_alloc_vnic_resources(enic);
1768         if (err) {
1769                 printk(KERN_ERR PFX
1770                         "Failed to alloc vNIC resources, aborting.\n");
1771                 goto err_out_free_vnic_resources;
1772         }
1773
1774         enic_init_vnic_resources(enic);
1775
1776         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1777          */
1778
1779         err = enic_set_nic_cfg(enic,
1780                 rss_default_cpu, rss_hash_type,
1781                 rss_hash_bits, rss_base_cpu,
1782                 rss_enable, tso_ipid_split_en,
1783                 ig_vlan_strip_en);
1784         if (err) {
1785                 printk(KERN_ERR PFX
1786                         "Failed to config nic, aborting.\n");
1787                 goto err_out_free_vnic_resources;
1788         }
1789
1790         /* Setup notification timer, HW reset task, and locks
1791          */
1792
1793         init_timer(&enic->notify_timer);
1794         enic->notify_timer.function = enic_notify_timer;
1795         enic->notify_timer.data = (unsigned long)enic;
1796
1797         INIT_WORK(&enic->reset, enic_reset);
1798
1799         for (i = 0; i < enic->wq_count; i++)
1800                 spin_lock_init(&enic->wq_lock[i]);
1801
1802         spin_lock_init(&enic->devcmd_lock);
1803
1804         /* Register net device
1805          */
1806
1807         enic->port_mtu = enic->config.mtu;
1808         (void)enic_change_mtu(netdev, enic->port_mtu);
1809
1810         err = enic_set_mac_addr(netdev, enic->mac_addr);
1811         if (err) {
1812                 printk(KERN_ERR PFX
1813                         "Invalid MAC address, aborting.\n");
1814                 goto err_out_free_vnic_resources;
1815         }
1816
1817         netdev->open = enic_open;
1818         netdev->stop = enic_stop;
1819         netdev->hard_start_xmit = enic_hard_start_xmit;
1820         netdev->get_stats = enic_get_stats;
1821         netdev->set_multicast_list = enic_set_multicast_list;
1822         netdev->change_mtu = enic_change_mtu;
1823         netdev->vlan_rx_register = enic_vlan_rx_register;
1824         netdev->vlan_rx_add_vid = enic_vlan_rx_add_vid;
1825         netdev->vlan_rx_kill_vid = enic_vlan_rx_kill_vid;
1826         netdev->tx_timeout = enic_tx_timeout;
1827         netdev->watchdog_timeo = 2 * HZ;
1828         netdev->ethtool_ops = &enic_ethtool_ops;
1829 #ifdef CONFIG_NET_POLL_CONTROLLER
1830         netdev->poll_controller = enic_poll_controller;
1831 #endif
1832
1833         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1834         default:
1835                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1836                 break;
1837         case VNIC_DEV_INTR_MODE_MSIX:
1838                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1839                 break;
1840         }
1841
1842         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
1843         if (ENIC_SETTING(enic, TXCSUM))
1844                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1845         if (ENIC_SETTING(enic, TSO))
1846                 netdev->features |= NETIF_F_TSO |
1847                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
1848         if (using_dac)
1849                 netdev->features |= NETIF_F_HIGHDMA;
1850
1851
1852         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
1853
1854         if (ENIC_SETTING(enic, LRO)) {
1855                 enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
1856                 enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
1857                 enic->lro_mgr.lro_arr = enic->lro_desc;
1858                 enic->lro_mgr.get_skb_header = enic_get_skb_header;
1859                 enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1860                 enic->lro_mgr.dev = netdev;
1861                 enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
1862                 enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1863         }
1864
1865         err = register_netdev(netdev);
1866         if (err) {
1867                 printk(KERN_ERR PFX
1868                         "Cannot register net device, aborting.\n");
1869                 goto err_out_free_vnic_resources;
1870         }
1871
1872         return 0;
1873
1874 err_out_free_vnic_resources:
1875         enic_free_vnic_resources(enic);
1876 err_out_dev_close:
1877         vnic_dev_close(enic->vdev);
1878 err_out_vnic_unregister:
1879         enic_clear_intr_mode(enic);
1880         vnic_dev_unregister(enic->vdev);
1881 err_out_iounmap:
1882         enic_iounmap(enic);
1883 err_out_release_regions:
1884         pci_release_regions(pdev);
1885 err_out_disable_device:
1886         pci_disable_device(pdev);
1887 err_out_free_netdev:
1888         pci_set_drvdata(pdev, NULL);
1889         free_netdev(netdev);
1890
1891         return err;
1892 }
1893
1894 static void __devexit enic_remove(struct pci_dev *pdev)
1895 {
1896         struct net_device *netdev = pci_get_drvdata(pdev);
1897
1898         if (netdev) {
1899                 struct enic *enic = netdev_priv(netdev);
1900
1901                 flush_scheduled_work();
1902                 unregister_netdev(netdev);
1903                 enic_free_vnic_resources(enic);
1904                 vnic_dev_close(enic->vdev);
1905                 enic_clear_intr_mode(enic);
1906                 vnic_dev_unregister(enic->vdev);
1907                 enic_iounmap(enic);
1908                 pci_release_regions(pdev);
1909                 pci_disable_device(pdev);
1910                 pci_set_drvdata(pdev, NULL);
1911                 free_netdev(netdev);
1912         }
1913 }
1914
1915 static struct pci_driver enic_driver = {
1916         .name = DRV_NAME,
1917         .id_table = enic_id_table,
1918         .probe = enic_probe,
1919         .remove = __devexit_p(enic_remove),
1920 };
1921
1922 static int __init enic_init_module(void)
1923 {
1924         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
1925
1926         return pci_register_driver(&enic_driver);
1927 }
1928
1929 static void __exit enic_cleanup_module(void)
1930 {
1931         pci_unregister_driver(&enic_driver);
1932 }
1933
1934 module_init(enic_init_module);
1935 module_exit(enic_cleanup_module);